With ggplot2, we dive into the world of programmatic data visualization. The ggplot2 library implements something called the grammar of graphics. The main concepts are:
Hadley Wickam, who is behind all of the libaries we have used in this course to date, wrote about his layered grammar of graphics in this 2009 paper that is worth your time to read.
Here are some ggplot2 resources you'll want to keep handy:
Let's dive in using data we've already seen before -- mountain lion sightings. This workflow will represent a clear picture of what your work in this class will be like for much of the rest of the semester. One way to think of this workflow is that your Jupyter Notebook is now your digital sketchbook, where you will try different types of visualizations to find ones that work. Then, you will export your work into a program like Illustrator to finish the work.
To begin, we'll import the ggplot2 and dplyr libraries. We'll read in the data, then create a new dataframe that represents our mountain lion data grouped by counties, similar to what we've done before.
In [1]:
library(ggplot2)
library(dplyr)
In [2]:
mountainlions <- read.csv("../../Data/mountainlions.csv")
The easiest thing we can do is create a simple histogram of our data. We could, for instance, create a bar chart of the number of times a county appears in the data. To do that, we simply tell ggplot2 what our dataset is, what element of the data we want to make the histogram out of (which is the aesthetic), and the geometry type. It looks like this:
ggplot(mountainlions, aes(x=COUNTY)) + geom_bar()
Note: mountainlions is our data, aes means aesthetics, x=COUNTY explicitly tells ggplot2 that our x value -- our horizontal value -- is the COUNTY field from the data, and then we add on the geom_bar() as the geometry. And what do we get when we run that?
In [3]:
ggplot(mountainlions, aes(x=COUNTY)) + geom_bar()
Well. We get a chart. So that's ... something. But it's kind of a mess. There's too many values, you can't read them on the bottom, they aren't in order. It's just a mess. So often times with bar charts, it's better to do our own grouping and counting, because we can control the data more than we can through ggplot2.
In [4]:
counties <- mountainlions %>%
group_by(COUNTY) %>%
summarise(
count = n(),
) %>% arrange(desc(count))
In [5]:
head(counties)
So what happens when we put counties in as our dataset, which did all the grouping and counting for ggplot2 instead of having ggplot do it?
In [6]:
ggplot(counties, aes(x=COUNTY)) + geom_bar()
Gah. That's awful. Everything is the same now! Well, ggplot2 default is to count each X value and use that to create the bar. So we grouped everything together, so there's only one of each county -- thus, why there's only one of everything. So we need to tell ggplot2 that we have something else to use to make the bars. In ggplot2, that's called the weight and it goes into the aesthetics.
In [7]:
ggplot(counties, aes(COUNTY, weight=count)) + geom_bar()
Better, but we're back where we started. What if we limited the chart to only the counties with a large number of sightings? We can do that in dpylr by using filter at the end. Note the last step of the data prep here:
In [8]:
counties <- mountainlions %>%
group_by(COUNTY) %>%
summarise(
count = n(),
) %>% arrange(desc(count)) %>%
filter(count > 10)
In [9]:
ggplot(counties, aes(COUNTY, weight=count)) + geom_bar()
Better still, but the bars aren't in any order that makes sense -- they're in alphabetical order by county, even though our dataset sorts them. That's because ggplot2 ignores any ordering you do to the data ahead of time, and the ordering done in arrange doens't really stick. To get it to do that, we have two options: We can reorder it in ggplot2 using reorder in the aesthetics -- reorder(COUNTY, count) which says reorder the order of the counties by the value of count.
In [10]:
ggplot(counties, aes(reorder(COUNTY, count), weight=count)) + geom_bar()
But that looks silly because the least value is first when it should be the greatest value. So we can change that pretty easily.
In [11]:
ggplot(counties, aes(reorder(COUNTY, -count), weight=count)) + geom_bar()
Our other option is to fix the ordering in our dplyr work with an additional mutate step. We do that by adding %>% mutate(COUNTY = factor(COUNTY, COUNTY)). What that says in English is mutate COUNTY by reordering the factor COUNTY by what is in COUNTY, which, thanks to arrange, is in the order we want it to be in.
In [12]:
counties <- mountainlions %>%
group_by(COUNTY) %>%
summarise(
count = n(),
) %>% arrange(desc(count)) %>%
filter(count > 10) %>% mutate(COUNTY = factor(COUNTY, COUNTY))
Which means we can now go back to our more simple ggplot2 code from before.
In [23]:
ggplot(counties, aes(COUNTY, weight=count)) + geom_bar() + ggtitle("Mountain Lion Sightings, 1994-2006") +
theme(plot.title = element_text(lineheight=.8, face="bold")) + theme_light()
One neat little trick? What if we didn't want vertical bars? We want horizontal bars? We can use coord_flip() to do that.
In [14]:
ggplot(counties, aes(COUNTY, weight=count)) + geom_bar() + coord_flip()
In [ ]: